a This paper proposes a dynamic Bayesian rolling window estimation procedure applied to the three-factor model of Fama and French to analyse herding behaviour in the style exposures of mutual funds. This procedure allows a user to dynamically select the length of the estimation window by means of weighted likelihood functions that discount the loss of information because of time. This method is very flexible and allows us to consider different approaches of detecting herding behaviour by taking into account the uncertainty associated in the estimation of the style coefficients. In particular, the paper first determines the convergence behaviour following the traditional LSV herding measure and then refines this method by removing the influence exerted by market conditions, such as market volatility and returns, on this convergence. This process is empirically illustrated by an application to Spanish equity mutual funds.